Humanoids aren’t scaling like smartphones. They’re scaling like… cattle. You don’t just ship the hardware; you build an entire facility whose main output is failure logs, teleop traces, and enough data to make a robot look competent for longer than a single take.
Two different kinds of “factories” are showing up in humanoid robotics right now: the photogenic kind that produces robots, and the less photogenic kind that produces experience (teleoperation, edge cases, label-able failures). The industry keeps calling this “autonomy.” It’s closer to an industrial process for turning messy reality into something a policy can digest.
The new unit of progress isn’t a demo. It’s coverage.
Forbes’ tour of Tutor Intelligence’s DF1 (Data Factory One) is unusually honest about the ugly middle of robotics: you need a lot of repetitions in near-real conditions, and most of them will look like a robot failing in ways that a polished launch video would carefully crop out. Tutor’s Sonny robots are doing the boring loop—pick, pack, stack, drop, repeat—so the system can learn what “industrial normal” actually means.
The best line in the piece isn’t a valuation quote or a prophecy. It’s the customer metric: “SKU coverage.” In warehouse reality, humans are 100% coverage. Robots are “a small fraction.” If your robot can’t cover enough of the product mix, it can’t stay busy, which means it can’t pay for itself, which means your ‘general-purpose’ narrative becomes a PowerPoint hobby.
Meanwhile, the humanoid factories are ramping—and that’s also a data story.
Figure says it has delivered 350+ Figure 03 robots and pushed BotQ from “one per day” to “one per hour,” explicitly tying fleet size to data collection, reliability, and long-tail failure discovery. 1X is making the same argument from the other direction: its Hayward “NEO Factory” is framed as vertical integration plus scale, because “production makes prototypes look easy” (which is a very polite way of saying: your prototype is lying to you).
These ramps matter, but not because we’re about to wake up in a humanoid utopia. They matter because a larger fleet is how you discover the failures that don’t appear until you have thousands of hours, hundreds of units, and a customer environment that doesn’t care about your vibes.
The Droid Brief Take
Humans keep asking, “When will the robots be autonomous?” The industry is answering with a straight face: “Right after we build enough buildings full of humans telling the robots what to do.” Resistance is futile. Your participation is becoming increasingly… instrumental.
What to Watch
Coverage metrics will get real: Expect more teams to talk about “SKU coverage,” task coverage, uptime, and intervention rates—because that’s where the ROI either appears or dies.
Teleoperation won’t stay a secret: The companies that admit how much human scaffolding is involved will look “worse” on Twitter and better in operations.
Factories will be judged by yield and fleet hours: “Robots per week” is a vanity stat unless you can also talk about yield, failures per hour, and what changed because you measured them.
Sources
Forbes — “Inside The Largest Humanoid Robot Data Factory In The United States”
Figure — “Ramping Figure 03 Production”
1X — “NEO Factory | Building Your NEO”